Constructing household specific consumer price indexes : an analysis of different techniques and methods

An earlier version of this paper was presented at the 1997 National Bureau of Economic Research Summer Institute on Price Index MeasurementThe primary purpose of this study is to produce household specific price indexes for consumer units or households living in the United States in the early 1990s. This paper is a report on how these household specific indexes were created. With household specific indexes, households are assumed to have nonhomothetic preferences, so changes in prices involve relative price changes between different sets of commodities and the resulting indexes will differ systematically between different households. We examine several different approaches to construct these indexes. Our indexes are based on internal U.S. Consumer Expenditure Survey (CEX) data for 1990-91 and Bureau of Labor Statistics Consumer Price Index (CPI) data from winter 1981, 1987, and 1991. Our base period is 1990-91. Using these data we produce Paasche type household specific indexes. In addition we propose an alternative definition of total expenditures, based on the CPI market basket commodity space, to be used for welfare analysis. Our underlying motivation for conducting this study was to compare real welfare inequality in Spain and the U.S. in the 1980s for another study (Garner et al. forthcoming 1997). Because of this comparison, we were somewhat restricted in our approach. CEX data are used to calculate CPI market basket item budget shares for each interviewed household. Price indexes are merged with the household budget data at various levels of geographic and market basket item aggregation, and the variability in these indexes are compared in order to measure the value of using detailed consumption space over aggregated consumption space. In this study we introduce two novel approaches to producing household specific price indexes using BLS data. First, expenditure data from the Consumer Expenditure Diary survey, which is more detailed than the Interview, are used to impute missing consumption items for the Interview households. And second, a method to impute household indexes for the rural population is presented. Two different types of samples, horizontal and vertical (based on assumptions about the Interview households selected to define the base period), are used to provide the weights for the price indexes. Indexes are presented based on Interview only items and all items commodity spaces for the horizontal and vertical samples with additional indexes produced for consumer units living in urban and rural areas. From our study we conclude that indexes based on expenditures for the horizontal and vertical samples do not differ significantly for the time periods of our study. However, differences in the indexes do result for the urban versus rural samples, with consumer units living in urban areas facing greater changes in relative prices than are faced by consumer units living in rural areas. The all-item indexes produced slightly higher index values than did the Interview-only item indexes. Relative prices appear to be pro-poor during the 1980s